tensorflow-yolov4-tflite | tiny Implemented in Tensorflow | Computer Vision library

Main Question:

What changes should I do to the repo's source code to successfully convert my YOLOv4 darknet weight (with custom anchors) to Tensorflow format?

Background:

I used this repo to convert my YOLOv4 darknet weights to Tensorflow format.

I have trained YOLOv4 on a custom dataset using custom anchors (9 anchors) but the number of anchors I used per [yolo] layer is 4, 3, 2, respectively. By default, YOLOv4 uses 3 anchors each [yolo] layer.

Main Problem:

The repo I used is coded in a way that only considers the default anchors, where there are 3 anchors each [yolo] layer.

What I tried to do to solve the main problem:

• I have tried to do some changes to the source code, which are summarized in this link.
• I used the code below to attempt converting the darknet weight to tf format. Here is the log of the conversion process.

python save_model.py --weights data/yolov4-512.weights --output ./checkpoints/yolov4-512 --input_size 512 --model yolov4

• I tested the resulting tf model using the code: python detect.py --weights checkpoints/yolov4-512 --size 512 --model yolov4 --image data/pear.jpg . The process failed and the error can be seen below. I have seen possible problems here but I don't know how to solve them.

How to convert YOLOv4-CSP darknet weights to Tensorflow (tf) format?

I have tried using this repo but it didn't work.

I had this error message:

I want to run yolov4 code in this repo: https://github.com/hunglc007/tensorflow-yolov4-tflite And I installed python 3.7 and all requirements and cuda and cudnn. By the log, the cudnn and cuda is installed well, but there is error of "no kernel image is available for execution on the device" what is this error? is it related in cuda or cudnn version error?

Python: 3.7.9, CUDA: 10.1, Tensorflow:2.3.0rc0, Tensorflow-GPU:not installed, CUDNN:7.5.0, OS: Windows10(x64)

My goal is to load a saved model once and use it for inference multiple times on different images to save time between each prediction. In my case, after loading the model, the first prediction is fine. However, if I try to use the model a second time, the result is empty. Is there a way to use the loaded model for inference multiple times or am I doing something terribly wrong?

I am using a trained YoloV4-tiny model that has been converted from a .wheights file to a .pb file using this repository. The tensorflow version I am using is tf-nightly 2.5.0.

Code sample:

I'm trying to use YOLO to detect license plate in an Android application.

So I train a YOLOv3 and a YOLOv4 model in Google Colab. I converted these 2 models to TensorFlow Lite, using the wonderfull project of Hunglc007 and I also verified that they are working and got the following result :

But when I try to understand the output of the model to adapt it in my app I got this using netron:

Why do I have 2 outputs when the model have been trained to detect only one single object?

And why the format of the output is like that, what does this [1,1,4] represents?

EDIT

The code for the bboxes can be found here